In article <2m6i7u$esc at mserv1.dl.ac.uk>
Jeroen Coppieters, jecop at gengenp.rug.ac.be writes:
>On 15 Mar 1994 j0m1742 at venus.tamu.edu wrote:
>>> These results indicate that even the fastest Power Mac
>> (8100/80) is barely faster in emulation mode than a
>> 40 MHz 030 chip (IIfx) while the 6100/60 is significantly slower
>> than the IIfx. The Quadra 840AV is the real peformer in
>> this group, over two times faster than any of the other
>>>> It is obvious that the Power Mac is not a good platform for
>> running PAUP unless you are using it mainly for other things
>> and are satisfied with IIfx-like performance. I understand that
>> David Swofford is working on a Unix version of PAUP and if that
>> is the case, PAUP users who will be analyzing large data sets
>> should probably save their money and wait for that option.
>This gives a completely wrong impression. I do not have PAUP here so I'm
>not sure about it. But I had a similar experience with ClustalV
>The reason of the slow performing PowerMac was that
>1)ClustalV is written in Think C (Is PAUP in Think C?)
>2) Think C has an it's own set of routines for emulating the
>coprocessor. If this is turned of, the code is not portable to Macs
>So, If you run Think C code that needs a coprocessor, it will be slow
>because you run an emulation of a 68040 with a disabled coprocessor.
>Recompile the program with switched coprocessor option and you will
>already see a big increase in performance, even though you still run the
>68040 emulation, and not a native compiled program.
I think both of these statements are a bit misleading. First of all, the
Power Mac has already amazed most people in its ability to run all but
a very few current software packages in the emulation mode. One shouldn't
expect better performance than a normal high-end 68030 Mac when running
in this mode. When Swofford gets around to recompiling the code with
a native ppc compiler, PAUP should fly! I would guess that it won't be
as difficult as some programs to port, because Swofford has already gone
to great extremes to make the basic functionality of PAUP portable to
all major platforms.
Second, PAUP depends very little on floating point calculations. It uses
integer bit operations almost exclusively. This means that a coprocessor
is not going to matter much at all. This is very different than maximum
likelihood software, for example, that is heavily floating point dependent.
The Power Mac should be very good at dealing with integer calculations,
as soon as the code is recompiled in native mode. How much of an increase
is gained for most programs is something that I have seen various estimates
of, but I suspect that it will be worth the effort in modifying the code.
PAUP is compiled for the Mac version in Think C, but there may be other
options (MetroWorks Code Warrier) for the conversion to Power Mac versions.